Local AI Changes Everything

GenAI Montréal @ Mirego Local AI Changes Everything GenAI Montréal is back at Mirego. *** \#\# Evening Schedule — Sept 8 Mirego — 18:00 PM Doors Open \#\#\# 6:00 PM — Arrival & Networking Meet members from GenAI Montréal \#\#\# ️ 6:30 PM — Welcome • Introduction to GenAI Montréal • Spotlight on next /dev/mtl Conference • Overview of the evening \#\#\# 6:45 PM — Talks Begin *** \#\# Speaker \#1 Jean-François René *Founder, Squid-OS* \#\#\# Bio Jean-François René began coding at seven on an Apple IIe and has spent the last three decades as a builder, founder, and technology leader. Over ten years as CTO at ScalePad, he helped grow the company from its early days to more than 300 people, establishing the product, platform, and engineering foundations behind its growth. A longtime open-source enthusiast, he is now returning to building from the ground up with Squid-OS, an open-source, local-first AI harness and agent runtime, while exploring how local AI is reshaping the way we interact with computers and build software. \#\#\# Talk A Journey to Local AI Abstract Running a capable AI model locally used to mean major compromises. That is changing quickly. In this talk, Jean-François René shares his hands-on journey into local AI: experimenting with open-weight models, pushing consumer hardware to its limits, navigating quantization, context windows, inference engines and tool calling—and discovering that the most interesting question isn't whether a local model can compete with the cloud. It's what becomes possible when the AI actually lives on your computer. The journey eventually led to Squid-OS, an open-source experiment in treating AI as an interface to the operating system itself. What started as an exercise in optimizing models shifted toward questioning the software we build around them: how low latency changes the interaction loop, whether we really need to consume so many tokens or hardcode so many capabilities into our harnesses, how small, smart and universal an AI harness can become, and how tools and reusable skills can extend models without constantly expanding their context. This is not a benchmark talk. It's a practical exploration of what works, what doesn't, what local AI still struggles with, and why increasingly capable local models may change how we build software and interact with our computers. As models become capable enough and harnesses become simpler, the boundaries between coding agents, OS agents and production agents start to disappear—and so does the hard boundary between local and cloud AI. *** \#\# Speaker \#2 Pascsal Hamel VP AI @ Mirego \#\#\# Bio Martin Gagnon is co-founder of Mirego and Senior Director, Artificial Intelligence, with over 25 years of experience in software development and technology leadership. He currently leads Mirego's AI Solutions team. In 2021, he launched Mirego's Technology Exploration team, which continuously builds prototypes evaluating emerging technologies and turns them, when relevant, into client-facing products. The team has delivered dozens of prototypes spanning AI, blockchain, and augmented reality. He has deep expertise in large language and vision models, as well as strong expertise in generative imaging with Stable Diffusion (including SDXL and Flux) and advanced workflows. He has built several intelligent assistants and agents, and contributed to the development of the Forra AI platform, enabling secure, private use of LLMs and the creation of assistants integrated with enterprise systems. \#\#\# Talk Agentic - Our Experience Building Intelligent Systems This talk shares our journey at Mirego building a range of agentic systems - what worked, what didn't, and what we learned along the way. This presentation is grounded in hands-on experience designing and deploying agentic systems in production, with a focus on practical lessons rather than pure theory. You'll leave with a clear understanding of when (and when not) to use these technologies —…